by Postpartum hemorrhage, a life-threatening condition and the leading cause of maternal mortality and morbidity globally, lacks comprehensive research and a universally agreed-upon definition. However, a recent study conducted by researchers from Brigham and Women’s Hospital has utilized Flan-T5, a large language model, to extract pertinent medical concepts from electronic health records. By doing so, they aim to enhance the understanding and identification of populations affected by postpartum hemorrhage.
The study demonstrated that the Flan-T5 model accurately identified patients with the condition in 95% of cases, leading to the identification of 47% more patients compared to the standard method that relies on billing codes for tracking. This tool holds immense potential in assisting clinicians in identifying subpopulations at a higher risk of postpartum hemorrhage and predicting which individuals are more likely to develop the condition.
The findings of this study have been published in the journal npj Digital Medicine.
Dr. Vesela Kovacheva, corresponding author of the study and affiliated with the Department of Anesthesiology, Perioperative and Pain Medicine, emphasized the need for improved methods to identify patients with postpartum hemorrhage and the various clinical factors associated with it. Dr. Kovacheva also expressed enthusiasm for the plethora of extraordinary large language models currently being developed, suggesting that this approach could be applied to other conditions and diseases.
The integration of artificial intelligence tools in healthcare has been revolutionary and possesses the potential to reshape the continuum of care in a positive manner.
Postpartum hemorrhage encompasses a wide range of patients, symptoms, and causes. To enable better categorization of patient subpopulations, the research team employed the Flan-T5 model to analyze comprehensive data from electronic health records.
The researchers stimulated the Flan-T5 model with lists of concepts known to be linked to postpartum hemorrhage and subsequently requested the model to extract these concepts from the discharge summaries of a cohort of 131,284 patients who had given birth at Mass General Brigham hospitals between 1998 and 2015. This method yielded rapid and accurate results without necessitating manual labeling.
Dr. Emily Alsentzer, the first author of the study and a research fellow in the Division, explained the approach utilized by the team. They cross-referenced the patients identified by Flan-T5 as having postpartum hemorrhage with those assigned the corresponding billing code. The results demonstrated a 95% accuracy rate and a 47% increase in the number of patients identified compared to relying solely on billing codes. Dr. Alsentzer highlighted the ultimate goal of predicting the occurrence of postpartum hemorrhage before it develops, expressing confidence in the efficacy of this tool in achieving that objective.
Moving forward, the research team aims to apply this approach to further investigate other pregnancy complications, ultimately hoping to address the rising maternal health crises in the United States.
Dr. Kovacheva emphasized the versatility of this approach, asserting its potential for application in future studies. Additionally, she highlighted its value in guiding real-time medical decision-making, which is particularly exciting and beneficial for clinicians.
*Note:
1. Source: Coherent Market Insights, Public sources, Desk research
2. We have leveraged AI tools to mine information and compile it
